ShieldaAI open-sources Agent Guard tool-call protection library
Understand this story
ShieldaAI has released the open-source TypeScript library Agent Guard (v1.3.1, Apache-2.0), used to decide which tool calls an AI agent may make. At the tool-call layer, it verifies whether a tool is registered, whether the task has been granted that resource, and whether the tenant and recipient are compliant, and it supports human approval and call budgets. The library targets Node.js 22 or later, with the executor and Agent Map requiring 22.13; the core has no third-party runtime dependencies. The included model adapter examples can run offline without an API key and cover the API formats of Anthropic, OpenAI Responses, OpenAI Chat Completions, Gemini, and Bedrock.
AI overview based on source reports · Updated 51 min ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIShow HN: Agent Guard – limit what an AI agent can do when it gets tricked
Agent Guard is a TypeScript library for Node.js 22+ that decides which tool calls an AI agent may make, checking each call against registered tools, task grants, tenant and recipient limits, per-action human approvals, and budgets. It assumes the agent can be misled and limits the effect rather than detecting prompt injection, and it does not cover actions taken outside the guard or sandbox tool handlers by default. Version 1.3.1 is Apache-2.0 licensed, with adapters for Claude, OpenAI, Gemini, Bedrock, the Vercel AI SDK, the OpenAI Agents SDK, LangChain and MCP clients.
Activity for this story
Not enough continuous observations to show a trend yet.